Professional Documents
Culture Documents
How To Create Your Own Virtual Machine! Part I
How To Create Your Own Virtual Machine! Part I
Part I
Presented by: Alan L. Bryan
A.k.a. Icemanind
Questions? Comments? Email me at
icemanind@yahoo.com
Please leave feedback if you enjoyed this tutorial. The more feedback I get, the more itll make me want to write Part II
Introduction
Welcome to my tutorial on virtual machines. This tutorial will introduce you to the concept of a
virtual machine and then we will, step by step, create our own simple virtual machine in C#. Keep in
mind that a virtual machine is a very complicated thing and even the simplest virtual machine can take
years for a team of programmers to create. With that said, dont expect to be able to create your own
language or virtual machine that will take over .NET or Java overnight.
In this tutorial, we will first layout the plan for our virtual machine. Then we will create a very
simple intermediate language. An intermediate language is the lowest level language still readable by
humans. It is comparable to assembly language, which is also the lowest level language on most
computers. The first program we will create will be a very simple intermediate compiler that will convert
our intermediate language to bytecode. Bytecode is a set of binary instructions that our virtual machine
will be able to directly execute. It is comparable to machine language, which is a set of binary or
machine instructions that all computers and CPUs understand. This virtual machine will be our second
project. It will be a virtual machine, created from scratch in C# that will execute our bytecode. It will be
very simple at first, but then we will expand it by adding threading support and dual screen outputs
(youll find out what Im talking about later).
All of the code in this tutorial is created using Visual Studio 2008 Professional, targeting the .NET
Framework 2.0. Since Im targeting the 2.0 framework, you should be able to use Visual Studio 2005 as
well. Since creating a virtual machine really does dive down into the nuts and bolts of how computers
work, I am assuming the reader of this has a pretty good, or a basic knowledge of, programming,
hexadecimal and binary number systems, and threading. It would also really help to know something
about assembly language, although I will try to help you understand things on a need-to-know basis.
If I havent scared you off and youre still interested in how to make a virtual machine, then lets
begin!
Planning it out
As described in the introduction, the first thing we will want to do is draw out a rough blue print
of what our machine will be able to do. I have decided to call our machine, B32 (Binary 32), although, for
simplicitys sake it will not be a 32-bit machine. It will be a 16-bit machine. B32 will have 64K of memory
and it can be addressed anywhere from $0000 - $FFFF. A B32 executable program can access any part of
that memory. Along with a 64K memory space, we will introduce 5 registers into our virtual machine. All
CPUs and all virtual machines have whats called registers. A register is similar to a variable. Registers
hold numbers and depending on how large the register is, determines how large of a number it can
hold. Unlike variables, however, registers do not take up memory space. Registers are built into CPUs.
This will make more sense once you see an example, which is coming up real soon.
To keep things simple, we will only implement 5 registers into our virtual machines. These
registers will be called A, B, D, X and Y. The A and B registers are only 8 bits in length, which means each
register can hold any number between 0 and 255 unsigned or between -128 to 127 signed. For now, we
are going to worry only about unsigned integers. We will get into signed later and we will briefly touch
on floating point numbers later. The X, Y and D registers will be 16 bits in length, capable of storing any
number between 0 and 65,535 unsigned or between -32768 to 32767 signed. The D register will be
something of a unique register. The D register will hold the concatenated values of the A and B registers.
In other words, if register A has $3C and register B has $10, than register D will contain $3C10. Anytime
a value in the A or B register is changed, then the value in the D register is also changed. The same is
true if a value in the D register is changed, the A and B registers will be changed accordingly. You will see
later why this is handy to have.
This has been a lot of dry talk, but here is a picture to represent our B32 registers:
B32 Registers
A
16
bits
16
bits
B
8
8
bits bits
D
16
bits
Hopefully this makes sense to you. If not, you will catch on as we progress through the tutorial.
Earlier when I told you that our virtual machine had 64K of free memory for an executable to
use, that was not entirely true. Really its only 60K because 4000 bytes must be reserved for screen
2009
2009
Mnemonic
Description
LDA
Assigns a value to our A register
$01
Example
LDA #$2A
LDX
$02
LDX #16000
STA
$03
END
$04
END START
You may be wondering what the pound sign # means in my examples above. The pound sign
will tell our assembler to use this value, that is, the value immediately following the pound sign. We
will introduce other forms of LDA, LDX and STA later in this tutorial, but for now, this is enough to get us
started.
For those of you who may also be wondering what the dollar sign $ means, it is a prefix that
will tell our assembler that the value is in hexadecimal format. If there is no dollar sign present, then the
assembler will assume the number is a regular integer number.
One final mnemonic that we will introduce is called END. This is not really a mnemonic
though. This is an assembler command that will tell our assembler this is the end of our program. All
B32 programs we created must have at least 1 and only 1 END statement and it should be the last line of
the program. The operand for our END statement will be a label that points to the part of our program
where execution will begin. We will discuss labels and execution points later in the tutorial.
One final piece of business we need to discuss before we get our hands dirty and start writing
our assembler is the file format of our B32 executables. To keep things simple, our file format will be as
follows:
Data
B32
Length
3 Bytes
<Starting Address>
2 Bytes
<Execution Address>
2 Bytes
<ByteCode>
?? Bytes
Description
Our magic header number
This is a 16-bit integer that tells our virtual machine where, in
memory, to place our program.
This is a 16-bit integer that tells our virtual machine where to begin
execution of our program.
This will be the start of our bytecode, which can be any length.
2009
Most binary file formats have a magic number as a header. A magic number is one or more
bytes that are unique to that file format. For example, all DOS and Windows executables start with
MZ. Java binary class files have 4 bytes for its magic number and start with $CAFEBABE. Our B32
executables will start with B32. There are two main purposes for this magic number. The first is, our
virtual machine can check to be sure the file its trying to execute is, indeed, a B32 binary file. The
second purpose for have magic numbers is some operating systems, such as Linux for example, can
automatically execute files by looking at this magic number in a database, then calling the appropriate
program.
B32 Assembler
Finally! Its time to get our hands dirty and start working on our assembler. The goal of the
assembler will be to translate our B32 mnemonics into a B32 binary. Our assembler is going to expect
input to be in the following format:
[Optional Label:]
<white space><mnemonic><white space><operand>[Optional white space]<newline>
A label starts with a letter and is composed of any number of letters or numbers, followed by a
colon. As far as the assemblers concerned, a label will simply be translated into a 16-bit value defining
an area of memory. A white space is any number of spaces or tabs. Each mnemonic MUST be preceded
by at least 1 space or 1 tab; otherwise, our assembler will treat the mnemonic as a label instead of as a
mnemonic. Likewise, each mnemonic must also have at least 1 space or 1 tab after the mnemonic. To
demonstrate this, we are going to create our very first B32 assembly language program right now. Open
up notepad or some other text editor of your choosing and type the following program EXACTLY as you
see it below and dont forget to put a single space before each mnemonic and also be sure to end the
last line with a newline:
START:
LDA #65
LDX #$A000
STA ,X
END START
Five points if you can guess what this program will do! Thats right! This program doesnt do
much except put a capital letter A in the upper left hand corner of the screen. The first line is our label.
This is where our execution will begin. The next line loads our A register with the value of 65. The ASCII
value of A is 65. The following line loads our X register with hex value $A000. If you remember from our
previous discussion, we said that our video memory will start at $A000, thus defining the upper left
2009
Control Type
Label
Label
Label
Label
TextBox
TextBox
TextBox
Button
Button
Button
OpenFileDialog
Name
Label1
Location
X = 16
Y = 23
Label2
X = 20
Y = 52
Label3
X = 44
Y = 77
Label4
X = 77
Y = 77
txtSourceFileName X = 87
Y = 20
txtOutputFileName X = 87
Y = 49
txtOrigin
X = 87
Y = 75
btnAssemble
X = 97
Y = 138
btnSourceBrowse
X = 193
Y = 17
btnOutputBrowse
X = 193
Y = 46
fdDestinationFile
N/A
Size
Autosize
Other properties
Text = Source File:
Autosize
Autosize
Text = Origin:
Autosize
Text = $
W = 100
H = 20
W = 100
H = 20
W = 100
H = 20
W = 75
H = 23
W = 75
H = 23
W = 75
H = 23
N/A
Text =
Text =
Text =
Text = Assemble!
Text = Browse
Text = Browse
Filter = B32 Files|*.B32
DefaultExt = B32
Filename =
2009
fdSourceFile
N/A
N/A
CheckFileExists = False
Filter
=
B32
Assembly
Files|*.asm
DefaultExt = asm
Filename =
Now double click the top browse button to create a click event handler for the button, then type in
the following code:
private void btnSourceBrowse_Click(object sender, EventArgs e)
{
this.fdSourceFile.ShowDialog();
this.txtSourceFileName.Text = fdSourceFile.FileName;
}
Now go back to designer view and double click the second browse button, then type in the following
code:
private void btnOutputBrowse_Click(object sender, EventArgs e)
{
this.fdDestinationFile.ShowDialog();
this.txtOutputFileName.Text = fdDestinationFile.FileName;
}
What this will do is allow the user to browse for a source and output file. If you run the program now
and click one of the browse buttons, it should pop up with a dialog box allowing you to find and choose
a file. Once you select a file and click OK, the filename should pop into the appropriate text box. The
origin will be where, in our 64K memory region, you want the program to be placed.
Now that we got our interface wired up, lets add the main functionality. Double click on the
Assemble! button to create a click event handler. Before coding the event handler though, add the
following class members to our class:
public partial class frmMainForm : Form
{
private string SourceProgram;
2009
System.Collections.Hashtable LabelTable;
int CurrentNdx;
ushort AsLength;
bool IsEnd;
ushort ExecutionAddress;
The SourceProgram variable will hold our program in memory. The B32 assembler will read
our source file and dump the contents into the SourceProgram variable. As described earlier,
LabelTable is a hash table that will hold our labels. The hash table will be populated during the first
stage of the assembly. CurrentNdx will be an integer variable that will be an index pointer to the
current location in the file. AsLength will be an unsigned 16-bit variable that will keep track of how big
our binary program is. IsEnd is simply a flag to determine if the end of the program has been reached.
Finally, ExecutionAddress will hold the value of our execution address. If some of this doesnt
make sense yet, it will as we code our program.
We are also going to need an enumeration that will store our registers. Add the following
enumeration just below the code you just entered:
private enum Registers
{
Unknown = 0,
A = 4,
B = 2,
D = 1,
X = 16,
Y = 8
}
We will put this enumeration to use later on when we start coding our helper functions. We will
create a function that will read a register from our program and return an enumeration type
representing the register.
Finally, before we start coding our Assemble button handler, add the following lines to the
frmMainForm class constructor. These lines will automatically initialize our variables we added earlier,
assign a default origin, and allocate memory for our hash table:
public frmMainForm()
{
InitializeComponent();
LabelTable = new System.Collections.Hashtable(50);
CurrentNdx = 0;
AsLength = 0;
ExecutionAddress = 0;
IsEnd = false;
SourceProgram = "";
txtOrigin.Text = "1000";
}
2009
The first thing we are doing is grabbing our origin address and converting it into an unsigned 16
bit value and assigning that to AsLength. Next we are creating a BinaryWriter stream for our output
and a TextReader stream for our input, then we are opening the output stream, creating the file if it
does not already exist or overwriting it if it does exist. Next, we are opening the source file and read the
entire contents and storing it into SourceProgram, then closing the input buffer. The next 3 lines
create the header and magic numbers for our B32 binary file format, as we discussed earlier. Also, as we
discussed earlier, our file header will contain the string B32, followed by the starting address and the
execution address. You can see that we writing the stringing address to the file, however you may be
confused as to why we are writing zero as the execution address. This simply serves as a placeholder for
now, since we do not yet know the execution address. We will come back to this spot after we parse the
source file and write the correct address. Next we call the Parse() function. We have not written this
function yet, so I will hold off on discussing the details for that function. Finally, as promised, we are
seeking back to the execution address and writing the correct address, then we close the buffers and we
are done! Pretty simple, huh? Well I am hiding a lot of details, so lets move on and discuss those details
in depth.
The Parse() function will also be a pretty simple function. It will simply scan our file for labels,
then scan our file again and compile it (2 phases, as discussed earlier):
10
2009
Pretty simple. It first resets the CurrentNdx to zero, then enters a loop that calls LabelScan() until the
end of the file has been reached. It then resets the IsEnd flag to false, CurrentNdx back to zero and
AsLength back to the starting address. Finally it starts on the second pass and actually writes the
bytecode for our output file.
Next, we need to write the code for our LabelScan() function:
private void LabelScan(System.IO.BinaryWriter OutputFile, bool
IsLabelScan)
{
if (char.IsLetter(SourceProgram[CurrentNdx]))
{
// Must be a label
if (IsLabelScan) LabelTable.Add(GetLabelName(), AsLength);
while (SourceProgram[CurrentNdx] != '\n')
CurrentNdx++;
CurrentNdx++;
return;
}
EatWhiteSpaces();
ReadMneumonic(OutputFile, IsLabelScan);
}
Our LabelScan() function starts out by checking the first character in the line. Remember earlier I told
you that in our source file, each mnemonic MUST be preceded with a space. This is why. Our assembler
looks at the first character and if its not a space, it assumes its a label. The program then decides if its
on pass 1 or pass 2 (determined by our IsLabelScan flag variable) and if its on pass 1, it adds the
label to our LabelTable hash table. The GetLabelName() function is one of many helper functions we
will create later. For now, just know that GetLabelName() will simply retrieve the name of the label.
After our assembler finds a label, it continues to basically eat characters till it finds a newline character
(because remember that after the label, there should be nothing else on the line). If our LabelScan()
function does not find a label, then it calls the EatWhiteSpaces() function (another helper function) to
eat the white spaces. It then calls the next function we are going to code, ReadMnemonic().
The ReadMnemonic() function does exactly what it sounds like. It reads in the next mnemonic
waiting to be read. This function is presented here now:
11
2009
The ReadMneumonic() function should be pretty self explanatory. It reads the mnemonic, then
compares it against several if statements. These if statements call various functions to interpret the
mnemonic. After interpreting the mnemonic, it eats characters till if finds the end of the line.
Of the three interpreting functions we have referenced so far, the first one I am going to
introduce here is the InterpretLDA() function:
private void InterpretLDA(System.IO.BinaryWriter OutputFile, bool
IsLabelScan)
{
EatWhiteSpaces();
if (SourceProgram[CurrentNdx] == '#')
{
CurrentNdx++;
byte val = ReadByteValue();
AsLength += 2;
if (!IsLabelScan)
{
OutputFile.Write((byte)0x01);
OutputFile.Write(val);
}
}
}
Again, this function is pretty simple to figure out. First it eats all white spaces. Then it checks to
see if the operand begins with a pound sign # (recall earlier, I said that whenever our assembler
encounters a pound sign in the operand, it means to use this value, or in better terms, you the value
12
2009
Similar to its counterpart InterpretLDA(), InterpretLDX() works almost exactly the same. The only
differences are, for one, we are reading a 16-bit word value instead of an 8-bit byte value. Second, we
incrementing our length by 3 instead of by 2 since the X register can load a 16-bit value. Also, notice we
are writing $02 instead of $01 since $02 is the bytecode assigned to LDX.
The next of the interpret functions is InterpretSTA():
private void InterpretSTA(System.IO.BinaryWriter OutputFile, bool
IsLabelScan)
{
EatWhiteSpaces();
if (SourceProgram[CurrentNdx] == ',')
{
Registers r;
byte opcode = 0x00;
CurrentNdx++;
EatWhiteSpaces();
r = ReadRegister();
switch (r)
{
case Registers.X:
opcode = 0x03;
break;
}
AsLength += 1;
13
2009
Remember our format for storing the value of the A register to a memory location pointed to by the X
register STA ,X. The first thing this function does is check for a comma. It then eats any white space
then calls the ReadRegister() function. This is a helper function that will return the appropriate register
enumeration. It then writes the bytecode and increments our AsLength variable by one.
The last of our interpret functions (for now) is DoEnd:
private void DoEnd(System.IO.BinaryWriter OutputFile, bool
IsLabelScan)
{
AsLength++;
if (!IsLabelScan)
{
OutputFile.Write((byte)0x04);
}
}
Again, this is pretty simple. It simply increments our AsLength variable and writes a $04 byte.
Our assembler is almost done. All we need to do now is define our helper functions. First, here is
the code to the ReadRegister() function:
private Registers ReadRegister()
{
Registers r = Registers.Unknown;
if ((SourceProgram[CurrentNdx]
(SourceProgram[CurrentNdx]
if ((SourceProgram[CurrentNdx]
(SourceProgram[CurrentNdx]
if ((SourceProgram[CurrentNdx]
(SourceProgram[CurrentNdx]
if ((SourceProgram[CurrentNdx]
(SourceProgram[CurrentNdx]
if ((SourceProgram[CurrentNdx]
(SourceProgram[CurrentNdx]
==
==
==
==
==
==
==
==
==
==
'X') ||
'x')) r
'Y') ||
'y')) r
'D') ||
'd')) r
'A') ||
'a')) r
'B') ||
'b')) r
= Registers.X;
= Registers.Y;
= Registers.D;
= Registers.A;
= Registers.B;
CurrentNdx++;
return r;
}
This function simply reads the next character in the input and returns the appropriate
enumeration. Simple enough to figure out how it works.
Next, we will define our ReadWordValue() function:
14
2009
This function is also pretty simple to figure out. It first checks for a dollar sign $. The dollar sign signals
to the function that the number about to be read in is a hexadecimal number and not an integer. It then
reads in the number and converts it to an unsigned short and returns the value.
The next function is the sister function to ReadWordValue(), called ReadByteValue():
private byte ReadByteValue()
{
byte val = 0;
bool IsHex = false;
string sval = "";
if (SourceProgram[CurrentNdx] == '$')
{
CurrentNdx++;
IsHex = true;
}
while (char.IsLetterOrDigit(SourceProgram[CurrentNdx]))
{
sval = sval + SourceProgram[CurrentNdx];
CurrentNdx++;
}
if (IsHex)
{
val = Convert.ToByte(sval, 16);
}
15
2009
The code is almost identical, except the final value is converted into a byte instead of an unsigned short.
The next helper function is our EatWhiteSpaces() function:
private void EatWhiteSpaces()
{
while (char.IsWhiteSpace(SourceProgram[CurrentNdx]))
{
CurrentNdx++;
}
}
This function is very easy to follow. It simply increments our source code pointer till it points to a
character thats not a white space.
Finally, the last helper function we need, GetLabelName():
private string GetLabelName()
{
string lblname = "";
while (char.IsLetterOrDigit(SourceProgram[CurrentNdx]))
{
if (SourceProgram[CurrentNdx] == ':')
{
CurrentNdx++;
break;
}
lblname = lblname + SourceProgram[CurrentNdx];
CurrentNdx++;
}
return lblname.ToUpper();
}
This function returns the name of the label and converts it to upper case (since B32 code is case
insensitive).
Congratulations! Our primitive assembler is now complete. Go ahead and try to run the
program. It should compile and run without any errors. For the source file, browse to the Test.ASM file
we created earlier, and then choose an appropriate output file. Click the Assemble! button and it
should assemble your program without any problems. After it is done, it should have created you a B32
executable file. Feel free to examine this file with a hex editor. If you do, you will notice that our B32
header is in there, along with 2 bytes for our starting address, 2 bytes for our execution address and the
16
2009
17
2009
As promised, I am first initializing the default screen location to $A000. Next, I am allocated 4,000 bytes
to our array and finally, I am initializing the values in the array. I am using 32 (which is ASCII for a blank
space) for the character and an attribute of 7. An attribute of 7 will produce gray text on a black
background, just like the old DOS computers used to do. Essentially, I am clearing the screen. Review our
discussion on attributes up above, if this is confusing.
Next, we are going to add a public method called Poke. Poke will load a value into the area
specified in memory, within the range of our screen address. It is presented here:
public void Poke(ushort Address, byte Value)
{
ushort MemLoc;
try
{
MemLoc = (ushort)(Address - m_ScreenMemoryLocation);
}
catch (Exception)
18
2009
This method first tries to create an offset and returns if there is an error (if there is an error, its because
the address being poked is not within our screen range), then it checks to make sure its in range, then
pokes to value into the memory location. Finally, the control refreshes itself to update.
Pokes sister method is called Peek. Peek will return the byte value stored in any video
memory location. Add this method to our program:
public byte Peek(ushort Address)
{
ushort MemLoc;
try
{
MemLoc = (ushort)(Address - m_ScreenMemoryLocation);
}
catch (Exception)
{
return (byte)0;
}
if (MemLoc < 0 || MemLoc > 3999)
return (byte)0;
return m_ScreenMemory[MemLoc];
}
This method is similar in the sense that it first tries to create an offset, validates the range, then returns
the appropriate byte.
The final thing we need to do to finish our B32Screen user control is to add code for the paint
handler. Switch back to designer mode, then create an event handler for the B32Screen Paint event. The
code for the paint handler is a little long, but after I present it, I will explain in detail how it works. Here
it is:
private void B32Screen_Paint(object sender, PaintEventArgs e)
{
Bitmap bmp = new Bitmap(this.Width, this.Height);
Graphics bmpGraphics = Graphics.FromImage(bmp);
Font f = new Font("Courier New", 10f, FontStyle.Bold);
int xLoc = 0;
int yLoc = 0;
for (int i = 0; i < 4000; i += 2)
19
2009
20
2009
21
2009
The code is quite long, but not really hard to figure out. The first thing Im doing is creating a bitmap
object to match the size of the control. In order to avoid flashing and/or flickering, I am creating the text
on a bitmap, and then transferring the bitmap to the screen. This will avoid all flashing and flickering
issues when a redraw() is needed. Next, I am creating a font for the text. I decided to use Courier New
for two reasons. One, almost all computers have a Courier font installed and two; it is a mono spaced
font. A mono spaced font is a font in which all the characters are the same width. In other words, if I
type a sentence thats 20 characters long, then type another sentence below it thats also 20 characters
long, the text will start and end at the same exact spot. This differs from a font you may use to type a
letter for example. If you got Microsoft Word, open it up and type 2 sentences below each other and
you will see the spacing is different. A capital O may take up more space than a lower case I for
example.
22
2009
What we are doing is poking 65 (A in ASCII) to memory location $A000, which is the start of our
screen memory. Go ahead and run the program and you should see an A in the upper left hand corner
in a light-gray color. If you do not, recheck the program for mistakes. If all works as expected, add the
following 2 lines:
public partial class MainForm :
{
public MainForm()
{
InitializeComponent();
b32Screen1.Poke(0xa000,
b32Screen1.Poke(0xa002,
b32Screen1.Poke(0xa004,
}
}
Form
65);
66);
67);
What I am doing is poking 66 (B in ASCII) and 67 (C in ASCII). When you run the program now, it
should display ABC in the upper left hand corner. Notice, in the program, that I am poking
incrementally by 2. Remember that the odd bytes (the ones Im not poking) control the attribute.
Speaking of which, modify the program as follows:
public partial class MainForm :
{
public MainForm()
{
InitializeComponent();
b32Screen1.Poke(0xa000,
b32Screen1.Poke(0xa002,
b32Screen1.Poke(0xa004,
Form
65);
66);
67);
23
2009
Run the program now and youll see our text is now bright white, housed inside a blue, red and green
background. Feel free to mess around and add some lines of your own. Write your own name, if you
wish. If you are confused as to how the colors are determined from the given binary pattern, refer to
the earlier discussion about attributes and colors and how they work.
Now that we tested our screen and we know it works, go ahead and remove all the lines we
added in the constructor (do not remove the InitializeComponent() call), along with any lines you may
have added. Switch back to design view. Add a menustrip control to MainForm. It should dock to the
top. This menu strip will provide us with a way to open a B32 bytecode file. Change the name of the
menu strip to msMainMenu. For now, just add a &File menu header to the strip and under that,
add &Open. It should like similar to the following:
Next, add a panel to the form. Change the dock to Bottom and change the name to
pnlRegisters. Make the height of the panel 54 pixels long. This panel will be a diagnostic panel that
will monitor the state of our registers and display the values of them accordingly. This way, you can see
that the B32 program we created is indeed executing as it should, without any tricks.
Drag an OpenFileDialog control onto the form. This dialog will popup when we click Open on our
menu. This will allow the user to search for and select a B32 binary file to be executed. Change the
24
2009
The first member field, B32Memory, will be our 64K memory, in the form of an array. We will initialize
that later, in the constructor. The StartAddr will contain the starting address of our B32 binary.
Likewise, the ExecAddr will be the execution address of our B32 binary. The InstructionPointer will
contain an address of where, in our 64K memory, that our next bytecode to be executed is at. The
remaining lines hold the values of our register.
Our constructor will initialize all our fields to default values:
public MainForm()
{
InitializeComponent();
B32Memory = new byte[65535];
StartAddr = 0;
ExecAddr = 0;
Register_A = 0;
Register_B = 0;
Register_D = 0;
Register_X = 0;
Register_Y = 0;
UpdateRegisterStatus();
}
Our B32Memory is initialized with a 64K byte array. Everything else is set to 0. UpdateRegisterStatus() is
a function we will code in a minute that will update the register label in our panel to display the
appropriate contents of our registers. Here is that function now:
25
2009
When this function is called, it takes the current values of the registers, converts them into hexadecimal,
then it sticks the text into our label. Should be pretty simple to follow this code. The PadLeft() functions
make the values consistent length, so that $01 is displayed instead of $1.
Switch back over to design view and create an event handler for our File Open menu item.
Type the following code into the event handler, then we will analyze it more in depth:
private void openToolStripMenuItem_Click(object sender, EventArgs e)
{
byte Magic1;
byte Magic2;
byte Magic3;
openFileDialog1.ShowDialog();
System.IO.BinaryReader br;
System.IO.FileStream fs = new
System.IO.FileStream(openFileDialog1.FileName, System.IO.FileMode.Open);
br = new System.IO.BinaryReader(fs);
Magic1 = br.ReadByte();
Magic2 = br.ReadByte();
Magic3 = br.ReadByte();
if (Magic1 != 'B' && Magic2 != '3' && Magic3 != '2')
{
MessageBox.Show("This is not a valid B32 file!", "Error!",
MessageBoxButtons.OK, MessageBoxIcon.Error);
return;
}
StartAddr = br.ReadUInt16();
ExecAddr = br.ReadUInt16();
ushort Counter = 0;
while ((br.PeekChar() != -1))
26
2009
The first thing Im doing is declaring 3 variables to hold our magic header numbers. Remember earlier,
we said that all B32 binaries will have a B32 as the magic header? I am then displaying our open file
dialog so that the user can select the B32 file. We then use a BinaryReader stream to open the file. We
then read in the file and check to make sure it has our B32 header. If it doesnt we display a message
box informing the user that this is not a valid file. We then read in the starting address and the
execution address, which is part of our B32 header. Next, we read in the bytecode and store it,
beginning at our start address. Finally, we close the stream, point our instruction pointer to the
execution address, then we call a function to execute our program.
Our ExecuteProgram() function does most of the real work. Here it is:
private void ExecuteProgram(ushort ExecAddr, ushort ProgLength)
{
ProgLength = 64000;
while (ProgLength > 0)
{
byte Instruction = B32Memory[InstructionPointer];
ProgLength--;
if (Instruction == 0x02) // LDX #<value>
{
Register_X = (ushort)((B32Memory[(InstructionPointer +
2)]) << 8);
Register_X += B32Memory[(InstructionPointer + 1)];
ProgLength -= 2;
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x01) // LDA #<value>
{
Register_A = B32Memory[(InstructionPointer + 1)];
SetRegisterD();
ProgLength -= 1;
InstructionPointer += 2;
UpdateRegisterStatus();
continue;
27
2009
The first think we do is read in an instruction bytecode. Then, using a series of if statements, we
execute that bytecode. After each bytecode is interpreted, we make a call to UpdateRegisterStatus()
to update our register label panel. We need only one last function to be able to execute and try out our
new virtual machine. Remember that whenever a value in the A register or the B register is modified,
the D register needs to also automatically update, and same thing if a value is modified in the D
register; we need to update the A and B registers accordingly. Here is the SetRegisterD() function that
updates the D register:
private void SetRegisterD()
{
Register_D = (ushort)(Register_A << 8 + Register_B);
}
This should be a pretty simple function to wrap your head around. I am pulling the value of the A
register, shifting it 8 bits to the left to get it into the upper significant bits, then I am adding the value of
the B register. Finally, I am casting the result to an unsigned short (16-bit value).
Go ahead and run our program now. Click on the File menu and go to open. Open the B32 test
file we assembled earlier. You should see an A appear in the upper left hand corner of the B32 screen,
along with the status of the registers at the bottom of the window! Congratulations! We have just
created a working virtual machine. Albeit, it is a pretty simple and limited virtual machine, but it does
work! And just to prove it, we are going to expand our test program from earlier. Exit the virtual
machine and open your test.asm file we made earlier. Change the file so it looks like the following:
Start:
LDA #65
LDX #$A000
STA ,X
LDA #66
LDX #$A002
STA ,X
28
2009
#67
#$A004
,X
Start
When typing this, dont forget to add at least one space before each mnemonic. Also, dont forget to
end the last line with a carriage return.
Open the B32 assembler and assemble Test.ASM, the new program you just created. Set the
output to Test.B32, then run the B32 virtual machine again, this time selecting the Test.B32 file we
just assembler. This time, it should display the letters ABC in the upper left hand corner. How cool is
this? Technically, you could write a B32 assembly file to do whatever you want, assuming you just use
the 4 mnemonics that we have programmed so far. This is pretty limiting though, and other then some
text being written to the screen, you couldnt create much of an application. We are going to spend the
remainder of this tutorial improving upon the virtual machine and assembler that we created here. For
now, if youd like, feel free to create some test ASM files and trying them out. Try to write your name,
centered, on the screen. Make it appear with a blue background and bright white text (hint: revisit the
attributes section we discussed earlier).
29
2009
Unused
Unused
Unused
Greater
Than
Less
Than
Not
Equal
Equal
1 Byte
Whenever we perform a compare operation, internally, each bit, or flag, gets set. Typically, either the
last bit will get set, indicating the register and the value compared against are equal, or else the not
equal flag will get set, along with one other flag, indicating if the register value was greater than or less
than the value compared against. Once we can do some sort comparison in B32, we can then cause
program execution to jump to another spot based on that comparison. Once we implement this, we can
start to do some more interesting stuff.
With all that said, our plan will be to implement the following mnemonics:
Mnemonic
CMPA
$05
Description
Example
Compares the value of the A CMPA #$20
register
CMPB
$06
CMPX
$07
CMPY
$08
CMPD
$09
30
2009
Description
Example
Jumps to a specific location JMP #$3000
in memory and resumes
execution
Jumps to a specific location CMPA #$6A
in memory ONLY if the result JEQ #$3000
of the last compare was
equal
JNE
$0C
JGT
$0D
JLT
$0E
This may seem like its going to be a lot of work to implement all these mnemonics, but youll soon see
that adding new mnemonics to our assembler and virtual machine is very easy to do.
Now that we have defined our goals and what we want to do, its time to make a test assembly
file that will test our new mnemonics. Open Test.asm in notepad again and delete all the lines and
type in the following for our test program (remember to put a space before each mnemonic and a
carriage return after the last line):
Start:
JMP #Spot1
LDA #65
LDX #$A000
STA ,X
JMP #EndSpot
Spot1:
LDA #72
LDX #$A002
STA ,X
CMPA #72
JEQ #Spot2
31
2009
32
2009
Pretty simple change. We test to see if the value is going to be hexadecimal. If its not and the current
character is not a number, we can safely assume its a label. We then access our LabelTable hash
table to retrieve the value and cast it to ushort.
33
2009
This should be pretty self explanatory. We are simply adding more if statements to test for the
presence of our jump and compare mnemonics. This works no differently then when we added LDA,
STA, END or LDX.
34
2009
35
2009
Each of these functions is almost identical. The only difference is the bytecode value that gets written.
There should be no mystery to understanding how each of these functions work. Now, here are the
jump functions:
private void InterpretJMP(System.IO.BinaryWriter OutputFile, bool
IsLabelScan)
{
EatWhiteSpaces();
if (SourceProgram[CurrentNdx] == '#')
{
CurrentNdx++;
AsLength += 3;
if (IsLabelScan) return;
ushort val = ReadWordValue();
if (!IsLabelScan)
{
OutputFile.Write((byte)0x0A);
OutputFile.Write(val);
}
}
}
private void InterpretJEQ(System.IO.BinaryWriter OutputFile, bool
IsLabelScan)
36
2009
37
2009
These are almost identical to our compare statements. The biggest difference is, notice that, if we are on
Pass 1, doing the label scan, we return early. The reason is, because the any forward labels we use will
not yet be defined, causing an error.
Thats it! See how simple it is to expand our assembler? We just added 10 brand new
mnemonics and it took all of 15 minutes to do it! Go ahead and run the assembler and try to assemble
the Test.ASM file you created earlier. It should we assemble just fine, producing an 82 byte B32
bytecode file. Now its time to expand our virtual machine so that it will be able to interpret the new
bytecodes.
38
2009
We define a byte to act as our compare flag and then we give it an initial value of 0 in the constructor.
The next change we need to make is to add lines to our ExecuteProgram() function. Scroll and find this
function and add the following new lines:
if (Instruction == 0x04) // END
{
InstructionPointer++;
UpdateRegisterStatus();
break;
}
if (Instruction == 0x05) // CMPA
{
byte CompValue=B32Memory[InstructionPointer+1];
CompareFlag = 0;
if (Register_A
(byte)(CompareFlag | 1);
if (Register_A
(byte)(CompareFlag | 2);
if (Register_A
(byte)(CompareFlag | 4);
if (Register_A
(byte)(CompareFlag | 8);
== CompValue) CompareFlag =
!= CompValue) CompareFlag =
< CompValue)
CompareFlag =
> CompValue)
CompareFlag =
InstructionPointer += 2;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x06) // CMPB
{
byte CompValue = B32Memory[InstructionPointer + 1];
CompareFlag = 0;
if (Register_B
(byte)(CompareFlag | 1);
if (Register_B
(byte)(CompareFlag | 2);
if (Register_B
(byte)(CompareFlag | 4);
if (Register_B
(byte)(CompareFlag | 8);
== CompValue) CompareFlag =
!= CompValue) CompareFlag =
< CompValue) CompareFlag =
> CompValue) CompareFlag =
InstructionPointer += 2;
39
2009
== CompValue) CompareFlag =
!= CompValue) CompareFlag =
< CompValue) CompareFlag =
> CompValue) CompareFlag =
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x08) //CMPY
{
ushort CompValue =
(ushort)((B32Memory[(InstructionPointer + 2)]) << 8);
CompValue += B32Memory[(InstructionPointer + 1)];
CompareFlag = 0;
if (Register_Y
(byte)(CompareFlag | 1);
if (Register_Y
(byte)(CompareFlag | 2);
if (Register_Y
(byte)(CompareFlag | 4);
if (Register_Y
(byte)(CompareFlag | 8);
== CompValue) CompareFlag =
!= CompValue) CompareFlag =
< CompValue) CompareFlag =
> CompValue) CompareFlag =
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x09) //CMPD
{
ushort CompValue =
(ushort)((B32Memory[(InstructionPointer + 2)]) << 8);
CompValue += B32Memory[(InstructionPointer + 1)];
CompareFlag = 0;
40
2009
== CompValue) CompareFlag =
!= CompValue) CompareFlag =
< CompValue) CompareFlag =
> CompValue) CompareFlag =
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
}
}
Most of these functions work identical to each other. The code should be pretty simple to follow. We
are setting the compare flag to zero each time a comparison is made, and then toggling the appropriate
bit depending on weather the comparison is equal, not equal, greater than, or less than.
Well if you thought that was easy, youre going to love the jump functions. Those are even
easier! Add the following lines to the ExecuteProgram() function:
if (Instruction == 0x09) //CMPD
{
ushort CompValue =
(ushort)((B32Memory[(InstructionPointer + 2)]) << 8);
CompValue += B32Memory[(InstructionPointer + 1)];
CompareFlag = 0;
if (Register_D
(byte)(CompareFlag | 1);
if (Register_D
(byte)(CompareFlag | 2);
if (Register_D
(byte)(CompareFlag | 4);
if (Register_D
(byte)(CompareFlag | 8);
== CompValue) CompareFlag =
!= CompValue) CompareFlag =
< CompValue) CompareFlag =
> CompValue) CompareFlag =
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x0A) // JMP
{
ushort JmpValue = (ushort)((B32Memory[(InstructionPointer
+ 2)]) << 8);
JmpValue += B32Memory[(InstructionPointer + 1)];
InstructionPointer = JmpValue;
41
2009
42
2009
Again, most of these functions work almost the save, jumping to an appropriate place in memory
depending on the bit state of our compare flags.
One last thing I want to do before we try out our new virtual machine. Scroll and find the
function we created earlier called UpdateRegisterStatus(). Add the following line:
strRegisters += "\nRegister X = $" +
Register_X.ToString("X").PadLeft(4, '0');
Register Y = $" +
strRegisters += "
Register_Y.ToString("X").PadLeft(4, '0');
strRegisters += "
Instruction Pointer = $" +
InstructionPointer.ToString("X").PadLeft(4, '0');
strRegisters += "\nCompare Flag = $" +
CompareFlag.ToString("X").PadLeft(2, '0');
this.lblRegisters.Text = strRegisters;
This will add a new line to our register status bar and show us the value of the compare flag. Go ahead
and run the virtual machine and then run out B32 bytecode binary. In the upper left hand corner of the
screen it should read HIJKL. Look again at the Test.asm file and you should be able to figure out
what is going on. Notice that Spot6 is never hit. If Spot6 had been hit, your display would say
HIJKLM instead of HIJKL. If you were to change that last JEQ #Spot6 to JNE #Spot6, reassembled and re-ran our program, then you would get HIJKLM as the output. Hopefully everything
up to this point makes sense. Do not continue on with this tutorial unless you have a firm understanding
of everything we learned up to this point. In the next section, we are going to implement some timing
functionality to our virtual machine.
43
2009
Timing is Everything!
Some virtual machines, especially ones that emulate an older machine from the 80s or early
90s, try to target a fixed clock cycle. Most modern day computers run at 2 GHz or 3 GHz or even 4 GHz,
while older machines only 2 MHz (thats MHz or Megahertz). If we wanted to emulate a 6502
microprocessor, for example, the way we created our virtual machine now, it would execute 6502
instruction codes WAY too fast. The way to properly introduce precision timing in our virtual machine
would be to find a hardware guide to the 6502 processor, find out how many clock cycles an instruction
will take to execute (which will probably be different for each instruction), then introduce some kind of
delay in between instruction execution commands.
For our virtual machine in this tutorial, we are not trying to target a specific clock cycle timing. I
am, however, demonstrating timing here because I think its important, as a learning tool, to see the
register values and flags change with each instruction. Our virtual machine does have a register bar that
we created; however, our program executes way too fast to see each and every change occur per
instruction. So in this section, we are going to improve our virtual machine so that we can adjust the
speed at which our virtual machine executes instructions. Its worth mentioning that the method we will
use here to delay is a static function found inside the System.Threading.Thread class. This function is
called Sleep() and is used to pause our program for XXX number of milliseconds. The Sleep() function is
defined as guaranteeing that our process will Sleep() for AT LEAST XXX number of milliseconds, however
it is considered inaccurate because it will not guarantee that our process will Sleep() for PRECISLY XXX
number of seconds. It is also not a good choice because sometimes youll want to delay for less than a
millisecond, which Sleep() does not support. Most real emulators will use some kind of high precision
and accurate timer, such as the timers found in DirectX. Since we are simply trying to learn about virtual
machines and not trying to actually make a full blown commercial product, Sleep() will suffice for us in
this tutorial.
To begin, open the B32Machine project inside Visual Studio, if you dont already have it open.
Switch over to designer view and we are going to add some more menu items. Add a menu item under
Open called Speed, and then under Speed, we are going to add 8 different speeds. We are going to
add a 1 second, 2 second, 3 second, 4 second, 5 second, second, second, and finally a Real Time
options, which will be a speed with no delay. Your menu should look like the following:
44
2009
If you have named the menu items exactly the same way as I named them, then the IDs should
be the same as used in the code below. Create a click handler for each of the 8 menu items you created
and place the following code in the appropriate handler:
private void mS14SecondToolStripMenuItem_Click(object sender,
EventArgs e)
{
UncheckAll();
((ToolStripMenuItem)sender).Checked = true;
SpeedMS = 250;
}
private void mS12SecondToolStripMenuItem_Click(object sender,
EventArgs e)
{
UncheckAll();
((ToolStripMenuItem)sender).Checked = true;
SpeedMS = 500;
}
private void mS1SecondToolStripMenuItem_Click(object sender,
EventArgs e)
{
UncheckAll();
((ToolStripMenuItem)sender).Checked = true;
SpeedMS = 1000;
}
private void mS2SecondsToolStripMenuItem_Click(object sender,
EventArgs e)
{
UncheckAll();
45
2009
Basically, each handler does almost the same thing. First, we uncheck all the menu items. Then we check
the menu item that was selected. Finally, we set the variable SpeedMS (which we did not create yet) to
the number of milliseconds we want to delay.
Go ahead and the SpeedMS field to the beginning of the class and initialize it in the constructor:
public partial class MainForm : Form
{
private byte[] B32Memory;
private ushort StartAddr;
private ushort ExecAddr;
private ushort InstructionPointer;
private byte Register_A;
private byte Register_B;
private ushort Register_X;
private ushort Register_Y;
private ushort Register_D;
private byte CompareFlag;
private ushort SpeedMS;
46
2009
We are simply defaulting it to 0 milliseconds (which is real time) and we are defaulting a check mark in
the RealTime menu strip, so that when we first run the program, this will be the default menu option
selected.
One more function we need to add before we can test the user interface:
private void UncheckAll()
{
mS12SecondToolStripMenuItem.Checked = false;
mS14SecondToolStripMenuItem.Checked = false;
mS1SecondToolStripMenuItem.Checked = false;
mS2SecondsToolStripMenuItem.Checked = false;
mS3SecondsToolStripMenuItem.Checked = false;
mS4SecondsToolStripMenuItem.Checked = false;
mS5SecondsToolStripMenuItem.Checked = false;
realTimeNoDelayToolStripMenuItem.Checked = false;
}
//
//
//
//
//
//
//
//
1/2 second
1/4 second
1 second
2 seconds
3 seconds
4 seconds
5 seconds
real time
This function un-checks all the menu items under the Speed menu. Go ahead and run the program now
and test the menu items out. Whenever you select a speed under the Speed menu item, it should
place a checkmark next to it and uncheck the previous one that was checked.
Now we will add the functionality needed to actually make the speed options work. Believe it or
not, this is as simple as adding just 1 line of code. Find the ExecuteProgram() function and the following
line of code:
private void ExecuteProgram(ushort ExecAddr, ushort ProgLength)
{
ProgLength = 64000;
while (ProgLength > 0)
{
byte Instruction = B32Memory[InstructionPointer];
ProgLength--;
System.Threading.Thread.Sleep(SpeedMS);
if (Instruction == 0x02) // LDX #<value>
As discussed earlier, this line makes a call to the System.Threading.Thread.Sleep static function. Go
ahead and run the program. Set the speed to 250 MS (1/4 of a second) and then open our B32 binary
test program that we used earlier. Notice how HIJKLM are slowly placed on the screen?
47
2009
48
2009
What we did here was first, comment out the line that calls ExecuteProgram(). We no longer want to
directly call this. Instead, on the next line, we created a new thread called prog and for the
construction parameters, we passed it a delegate to our ExecuteProgram() function and also passed
along the appropriate parameters to the function. Finally, the last new line we added starts the thread.
Believe it or not, we are actually almost done. We need to add three short functions to our
program:
private void ThreadPoke(ushort Addr, byte value)
{
if (b32Screen1.InvokeRequired)
{
PokeCallBack pcb = new PokeCallBack(Poke);
this.Invoke(pcb, new object[] { Addr, value });
}
else
{
Poke(Addr, value);
}
}
private void Poke(ushort Addr, byte value)
{
lock (b32Screen1)
{
b32Screen1.Poke(Addr, value);
}
}
private void SetRegisterText(string text)
{
lblRegisters.Text = text;
}
49
2009
What we did was comment out the old line and added our call to ThreadPoke(). The final change is to
the UpdateRegisterStatus() function and its shown here:
private void UpdateRegisterStatus()
{
string strRegisters = "";
strRegisters = "Register A = $" +
Register_A.ToString("X").PadLeft(2, '0');
strRegisters += "
Register B = $" +
Register_B.ToString("X").PadLeft(2, '0');
Register D = $" +
strRegisters += "
Register_D.ToString("X").PadLeft(4, '0');
strRegisters += "\nRegister X = $" +
Register_X.ToString("X").PadLeft(4, '0');
Register Y = $" +
strRegisters += "
Register_Y.ToString("X").PadLeft(4, '0');
Instruction Pointer = $" +
strRegisters += "
InstructionPointer.ToString("X").PadLeft(4, '0');
strRegisters += "\nCompare Flag = $" +
CompareFlag.ToString("X").PadLeft(2, '0');
50
2009
We are doing the same thing here that we did in ThreadPoke(). We are simple checking InvokeRequired
to see if we need to do an invoke and if so, we are creating a delegate to SetRegisterText().
Go ahead and build and run our solution now. Choose a speed of 250 milliseconds or 500
milliseconds and run our test B32 bytecode program. Notice the register bar accurately updates as the
program is running. Also notice that our interface doesnt freeze. We can move the window around, pull
down our File menu and even change the speed of the program dynamically, as its running.
One annoyance you may have noticed earlier on when we first developed our B32 virtual
machine is that each time we go to open a B32 bytecode program and run it, the screen does not reset.
Essentially, the data in our 64K memory retains its values every time we load a new program or the
same program. I would like to fix this annoyance now.
Bring up the code for the B32Screen control. Make the following changes to the constructor and
add the following function:
public B32Screen()
{
InitializeComponent();
m_ScreenMemoryLocation = 0xA000;
m_ScreenMemory = new byte[4000];
//for (int i = 0; i < 4000; i += 2)
//{
//
m_ScreenMemory[i] = 32;
//
m_ScreenMemory[i + 1] = 7;
//}
Reset();
}
public void Reset()
{
for (int i = 0; i < 4000; i += 2)
{
m_ScreenMemory[i] = 32;
m_ScreenMemory[i + 1] = 7;
}
Refresh();
}
51
2009
I added a local variable called dr. This variable will determine if the user canceled or closed the dialog
box. If so, we just return. Otherwise, we lock the b32Screen1 object (just in case the user clicked open
while a B32 binary was running) and perform a Reset().
Go ahead and run the program. Run our test B32 bytecode file, then change the speed to
something like 3000 milliseconds, then run our test file again. You will notice, this time, it does, indeed
reset the screen for us.
There is one final piece of business Id like to do before I close out our discussion of
multithreading. I want to implement a Restart and a Pause/Resume feature to our virtual machine.
Since our program is threaded now, it makes it a lot easier to implement these sort of functions.
Earlier we defined a local variable called prog. This variable was a System.Threading.Thread
object that executed our program. We need to change the scope of this variable so that it can be
accessed from anywhere in the class. So to do that, add the following line to the member declaration
section at the top of the class:
public partial class MainForm : Form
{
private byte[] B32Memory;
private ushort StartAddr;
private ushort ExecAddr;
private ushort InstructionPointer;
private byte Register_A;
private byte Register_B;
private ushort Register_X;
private ushort Register_Y;
52
2009
Finally, comment out our declaration from earlier and add the following line to the File Open
event click handler:
br.Close();
fs.Close();
InstructionPointer = ExecAddr;
//ExecuteProgram(ExecAddr, Counter);
//System.Threading.Thread prog = new
System.Threading.Thread(delegate() { ExecuteProgram(ExecAddr, Counter); });
prog = new System.Threading.Thread(delegate() {
ExecuteProgram(ExecAddr, Counter); });
prog.Start();
}
Okay, running and executing the program now should do exactly what it did before. The only change is
the thread is now class-wide.
Even though all Thread objects have methods for pausing and resuming thread executing
(Suspend(), Resume()), it is not a good idea to use these. Microsoft provided us with them for the
purpose of debugging and not to use on production code. The reason these shouldnt be used is because
there is no way of knowing what the process is doing at the time of suspension. If the thread was
processing a block of code inside one of our lock blocks, then the variable locked because suspended
indefinitely, causing whats called a dead lock.
Because of this, we are going to approach this a different way. Create a new instance member at
the top of the class:
public partial class MainForm : Form
53
2009
byte[] B32Memory;
ushort StartAddr;
ushort ExecAddr;
ushort InstructionPointer;
byte Register_A;
byte Register_B;
ushort Register_X;
ushort Register_Y;
ushort Register_D;
byte CompareFlag;
ushort SpeedMS;
Our thread will check the state of this PauseEvent and if its set, our program will wait indefinitely until
its reset. While you are near the top of the class, add the following two lines to the constructor:
public MainForm()
{
InitializeComponent();
prog = null;
CompareFlag = 0;
SpeedMS = 0;
realTimeNoDelayToolStripMenuItem.Checked = true;
resumeProgramToolStripMenuItem.Enabled = false;
pauseProgramToolStripMenuItem.Enabled = true;
B32Memory = new byte[65535];
StartAddr = 0;
These 2 lines will default the Resume menu item to disabled (since there is no program running
currently paused when we first run our virtual machine) and the Pause menu item to enabled.
The next change we need to make is in the click event handler for our File Open menu item.
Go to the bottom of the function and add the following line of code:
br.Close();
fs.Close();
InstructionPointer = ExecAddr;
//ExecuteProgram(ExecAddr, Counter);
//System.Threading.Thread prog = new
System.Threading.Thread(delegate() { ExecuteProgram(ExecAddr, Counter); });
prog = new System.Threading.Thread(delegate() {
ExecuteProgram(ExecAddr, Counter); });
PauseEvent = new System.Threading.ManualResetEvent(true);
prog.Start();
}
54
2009
The WaitOne() method will pause indefinitely, if our PauseEvent is reset. Create a click event handler for
our Pause menu item and add the following code to it:
private void pauseProgramToolStripMenuItem_Click(object sender,
EventArgs e)
{
resumeProgramToolStripMenuItem.Enabled = true;
pauseProgramToolStripMenuItem.Enabled = false;
PauseEvent.Reset();
}
Now create a click event handler for our Resume menu item and add the following:
private void resumeProgramToolStripMenuItem_Click(object sender,
EventArgs e)
{
resumeProgramToolStripMenuItem.Enabled = false;
pauseProgramToolStripMenuItem.Enabled = true;
PauseEvent.Set();
}
Go ahead and run the program. Set the speed to 3000 milliseconds, then open our test B32 bytecode
file. As its running, go ahead and try to pause our program.
The last thing we need to do is get our Restart working. Create a click handler for our
Restart click event and add the following code:
private void restartProgramToolStripMenuItem_Click(object sender,
EventArgs e)
{
if (prog != null)
55
2009
The first thing we do here is check to see if there is already a thread running. If so, we abort it and set
prog to null. Next, we point our InstructionPointer to the beginning of the execution address. Finally,
we create a new thread and a new PauseEvent and then we reset the B32 Screen and start the thread.
Go ahead and run the program again and open our test B32 file. After it executes, you should be
able to restart it as many times as you wish. This wraps up our multithreaded discussion. As we expand
our virtual machine, we will do it, keeping in mind the stuff we learned here. The next section, we will
continue to improve our assembler and virtual machine by adding some Pseudo-mnemonics and a
couple more regular mnemonics as well as some flags.
If you set a breakpoint and follow the value of j, you will see that it is first assign the value 255. After
the next line is hit, the value of j is 4. We had to put this in an unchecked code block, otherwise .NET
56
2009
2009
Left Shift
Carry
Right Shift
Carry
Carry Overf ow
Unused
57
Description
Increment the value in the A
register by 1
Increment the value in the B
register by 1
Increment the value in the X
register by 1
Increment the value in the Y
register by 1
Increment the value in the D
register by 1
Decrement the value in the A
register by 1
Decrement the value in the B
register by 1
Decrement the value in the X
register by 1
Decrement the value in the Y
register by 1
Decrement the value in the D
register by 1
Rotate A register to the left
Example
INCA
ROLA
ROLB
RORA
RORB
ADDB
$20
ADDAB
$21
LDB
$22
LDY
$23
INCB
INCX
INCY
INCD
DECA
DECB
DECX
DECY
DECD
ADDB #$30
58
2009
This looks like a lot new stuff, but really its only 3 different functions repeated for all the registers.
Notice I added an LDY and a LDB mnemonic. We created an LDA mnemonic earlier, but failed to
make an LDB mnemonic or LDY mnemonic. Truth be told, we had no use for it till now.
Open Test.ASM in notepad and type the following program (remember to precede each
mnemonic with a space and end the last line with a carriage return):
Start:
LDX #$A000
LDY #8
LDA #48
LDB #$81
Loop1:
ROLB
ADCA
STA ,X
LDA #48
INCX
INCX
DECY
CMPY #$00
JNE #Loop1
END Start
Big, Big points if you can figure out what this program will do. It might stump you at first, but once I
explain in, it should make sense. What it will do is output the 8-bit binary equivalent of the number
loaded in the B register. How it works is, it starts by rotating the B register to the left. The left most bit
is dropped into the carry flag. Register A is preloaded with 48. This is ASCII for 0. ASCII for 1 is 49.
So after the rotate, we are adding the value in the A register (which is 48) with the value in the carry
flag. If the carry flag is 0, then register A stays 48. If the carry flag is 1, then register A becomes 49.
This value is then outputted to the screen. We then increment the X register twice (remember twice,
not once to skip the attribute byte). We then decrement the Y register and check to see if its 0 yet. If
not, it loops back to our Loop1 label and repeats the process till all 8 bits are printed.
Time to implement these new mnemonics. Open Visual Studio and open your B32Assembler
project. The easiest ones to implement will be LDB and LDY, so lets do those first. Add the following
two lines to the ReadMneumonic() function:
if (Mneumonic.ToUpper() == "JNE") InterpretJNE(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "JGT") InterpretJGT(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "JLT") InterpretJLT(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "LDY") InterpretLDY(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "LDB") InterpretLDB(OutputFile,
IsLabelScan);
59
2009
It should be no mystery how these work. They work identical to InterpretLDA() and InterpretLDX(). In
fact, the only change that is different between the functions is the bytecode value thats written.
Next, we are going to write the increment and decrement code. Add the following lines to the
ReadMneumonic() function:
if
IsLabelScan);
if
IsLabelScan);
if
IsLabelScan);
if
IsLabelScan);
60
2009
61
2009
62
2009
All these function are pretty simple and they all work the same way. They write the bytecode, increment
the AsLength counter and thats it.
We will now finish up by adding the last of our mnemonics. Add the following lines to the
ReadMneumonic() function:
if (Mneumonic.ToUpper() == "DECB") InterpretDECB(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "DECX") InterpretDECX(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "DECY") InterpretDECY(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "DECD") InterpretDECD(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ROLA") InterpretROLA(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ROLB") InterpretROLB(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "RORA") InterpretRORA(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "RORB") InterpretRORB(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ADCA") InterpretADCA(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ADCB") InterpretADCB(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ADDA") InterpretADDA(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ADDB") InterpretADDB(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "ADDAB") InterpretADDAB(OutputFile,
IsLabelScan);
if (Mneumonic.ToUpper() == "END") { IsEnd = true;
DoEnd(OutputFile, IsLabelScan); EatWhiteSpaces(); ExecutionAddress =
(ushort)LabelTable[(GetLabelName())]; return; }
63
2009
64
2009
Go ahead and run the assembler and assemble our new Test.asm file. It should assemble without any
issues.
65
2009
Just like we did with the assembler, we will implement the LDB and LDY mnemonics first. Add
the following lines to the ExecuteProgram() function:
if (Instruction == 0x02) // LDX #<value>
{
Register_X = (ushort)((B32Memory[(InstructionPointer +
2)]) << 8);
Register_X += B32Memory[(InstructionPointer + 1)];
ProgLength -= 2;
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x23) // LDY #<value>
{
Register_Y = (ushort)((B32Memory[(InstructionPointer +
2)]) << 8);
Register_Y += B32Memory[(InstructionPointer + 1)];
ProgLength -= 2;
InstructionPointer += 3;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x01) // LDA #<value>
{
Register_A = B32Memory[(InstructionPointer + 1)];
SetRegisterD();
ProgLength -= 1;
66
2009
These functions work identical to their counterparts, LDA and LDX. Next, we are going to implement
the increment and decrement functions. These are easy and the code should be pretty straightforward.
Add the following lines to the ExecuteProgram() function:
if (Instruction == 0x0E) // JLT
{
ushort JmpValue = (ushort)((B32Memory[(InstructionPointer
+ 2)]) << 8);
JmpValue += B32Memory[(InstructionPointer + 1)];
if ((CompareFlag & 4) == 4)
{
InstructionPointer = JmpValue;
}
else
{
InstructionPointer += 3;
}
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x0F) // INCA
{
if (Register_A == 0xFF)
{
ProcessorFlags = (byte)(ProcessorFlags | 1);
}
else
{
ProcessorFlags = (byte)(ProcessorFlags & 0xFE);
}
unchecked { Register_A++; }
SetRegisterD();
InstructionPointer++;
67
2009
68
2009
69
2009
These should be pretty straight forward functions. We are adding 1 for the increment functions and
subtracting 1 for the decrement functions and setting our new flags appropriately. Next, we are going to
add our rotate functions. Add these functions just below where you added the functions above:
if (Instruction == 0x19) // ROLA
{
byte OldCarryFlag = (byte)(ProcessorFlags & 2);
if ((Register_A & 128) == 128)
{
ProcessorFlags = (byte)(ProcessorFlags | 2);
}
else
{
ProcessorFlags = (byte)(ProcessorFlags & 0xFD);
}
Register_A = (byte)(Register_A << 1);
if (OldCarryFlag > 0)
{
Register_A = (byte)(Register_A | 1);
}
SetRegisterD();
InstructionPointer++;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x1A) // ROLB
{
byte OldCarryFlag = (byte)(ProcessorFlags & 2);
70
2009
71
2009
To make these functions work, first we are saving the carry bit status to a temporary local variable. Then
we are setting the carry bit, according to either the leftmost bit or rightmost bit (weather its a left
rotate or a right rotate). We are then shifting the bits and finally, setting the appropriate bit to a 1 or 0,
which we got from the old carry flag.
Only thing left now is to add the functions that perform arithmetic addition. Add these functions
just below the functions you just added:
if (Instruction == 0x1D) // ADCA
{
if ((byte)(ProcessorFlags & 2) == 2)
{
if (Register_A == 0xFF)
{
ProcessorFlags = (byte)(ProcessorFlags | 1);
}
else
{
ProcessorFlags = (byte)(ProcessorFlags & 0xFE);
}
unchecked { Register_A++; }
SetRegisterD();
}
InstructionPointer++;
UpdateRegisterStatus();
continue;
}
if (Instruction == 0x1E) // ADCB
{
if ((byte)(ProcessorFlags & 2) == 2)
{
if (Register_B == 0xFF)
{
ProcessorFlags = (byte)(ProcessorFlags | 1);
}
else
{
ProcessorFlags = (byte)(ProcessorFlags & 0xFE);
}
72
2009
73
2009
Go ahead and run the program. Open the test file we made and run it. It should print an 8-bit binary
number to the screen. This is the binary representation of the hexadecimal number we loaded in the B
register. Feel free to change this number, re-assemble, than rerun our binary against it.
A Fond Farewell
I am closing out this document here. I am writing a part 2 that will pick up from this point exactly
and continue on. I hope you enjoyed this tutorial and you learned something from it. There is sooooo
much that can be added on. More mnemonics, keyboard input, graphics and/or multimedia extensions,
etc. can all be added to this. I would love to see examples of B32 programs. Creative programs. Feel free
to write me about questions or comments at icemanind@yahoo.com. I always look forward to emails.
74
2009